The Robust Vessel Segmentation and Centerline Extraction: One-Stage Deep Learning Approach.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Machine Learning in Healthcare

Background:

• Accurate blood vessel segmentation and centerline extraction are crucial for vascular imaging applications like surgical planning and hemodynamic modeling.
• Existing methods often require multiple stages or post-processing, impacting efficiency and accuracy.
• There is a need for integrated, end-to-end solutions that directly address both segmentation and centerline extraction.

Purpose of the Study:

• To introduce a novel one-stage multitask neural network for simultaneous vessel segmentation and centerline extraction.
• To develop an end-to-end framework that directly predicts centerlines as connected polylines.
• To evaluate the performance and robustness of the proposed method against state-of-the-art techniques.

Main Methods:

• A hybrid neural network architecture integrating convolutional and graph layers was designed.
• A task-specific loss function was developed to leverage complementary features between segmentation and centerline extraction.
• The model was trained and evaluated on a combined dataset of 142 computed tomography angiography images (LIDC-IDRI and AMOS).

Main Results:

• The proposed method achieved superior centerline extraction performance (Surface Dice with threshold of 3 mm: 97.65%±2.07%).
• The highest subvoxel resolution was attained (Surface Dice with threshold of 1 mm: 72.52%±8.96%).
• The model demonstrated robustness against small rigid and deformable transformations and was benchmarked against the VMTK toolkit.

Conclusions:

• The novel one-stage method effectively performs simultaneous vessel segmentation and centerline extraction.
• The end-to-end framework eliminates the need for post-processing, improving efficiency.
• The approach offers superior accuracy and subvoxel resolution for vascular imaging analysis.